Objet : Developers list for StarPU
Archives de la liste
- From: Alfredo Buttari <alfredo.buttari@enseeiht.fr>
- To: starpu-devel@lists.gforge.inria.fr, Emmanuel Agullo <Emmanuel.Agullo@inria.fr>, Abdou Guermouche <abdou.guermouche@labri.fr>
- Subject: [Starpu-devel] huge regression in LWS scheduler
- Date: Wed, 8 Jun 2016 15:20:01 +0200
- Authentication-results: mail2-smtp-roc.national.inria.fr; spf=None smtp.pra=alfredo.buttari@gmail.com; spf=Pass smtp.mailfrom=alfredo.buttari@gmail.com; spf=None smtp.helo=postmaster@mail-wm0-f42.google.com
- Ironport-phdr: 9a23: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
- List-archive: <http://lists.gforge.inria.fr/pipermail/starpu-devel/>
- List-id: "Developers list. For discussion of new features, code changes, etc." <starpu-devel.lists.gforge.inria.fr>
Dear all,
I am making a few tests with qr_mumps on an Intel KNL machine with 64 cores and I am getting very weird performance with the latest trunk version when the LWS scheduler is chosen. Here are the results I am getting (exactly the same problem for every test, always using 64 cores):
starpu-1.2.0rc5 eager : 513 Gflop/s
starpu-1.2.0rc5 lws : 623 Gflop/s
starpu r17762 eager : 625 Gflop/s
starpu r17762 lws : 116 Gflop/s
I don't have exclusive access to the machine and it's not easy to run many tests but I can do a few more if you want.
I am making a few tests with qr_mumps on an Intel KNL machine with 64 cores and I am getting very weird performance with the latest trunk version when the LWS scheduler is chosen. Here are the results I am getting (exactly the same problem for every test, always using 64 cores):
starpu-1.2.0rc5 eager : 513 Gflop/s
starpu-1.2.0rc5 lws : 623 Gflop/s
starpu r17762 eager : 625 Gflop/s
starpu r17762 lws : 116 Gflop/s
I don't have exclusive access to the machine and it's not easy to run many tests but I can do a few more if you want.
Ciao,
Alfredo
--
-----------------------------------------
Alfredo Buttari, PhD
CNRS-IRIT
2 rue Camichel, 31071 Toulouse, France
http://buttari.perso.enseeiht.fr
--
-----------------------------------------
Alfredo Buttari, PhD
CNRS-IRIT
2 rue Camichel, 31071 Toulouse, France
http://buttari.perso.enseeiht.fr
- [Starpu-devel] huge regression in LWS scheduler, Alfredo Buttari, 08/06/2016
- Re: [Starpu-devel] huge regression in LWS scheduler, Samuel Thibault, 08/06/2016
- Re: [Starpu-devel] huge regression in LWS scheduler, Alfredo Buttari, 09/06/2016
- Re: [Starpu-devel] huge regression in LWS scheduler, Samuel Thibault, 08/06/2016
Archives gérées par MHonArc 2.6.19+.